This proposal addresses several of the bottlenecks in the process of refining macromolecular crystal structures. First, a new approaches to the identification of model errors is developed that automatically detects inappropriate parameterizations and explicitly shows the precision of determination of parameters. This approach, which depends on diagonalization of the refinement matrix, works even at less than atomic resolution. Second, scalable computing methods for refinement will be developed which take full advantage of modern parallel computing architectures, including inexpensive high-performance clusters. Third, analysis of the full second order refinement methods will allow explicit evaluation of computational optimizations such as sparse matrix approximations, which offer the prospect of rapid convergence at less cost. These advantages will be realized through the development of a modern, object-oriented refinement program [unreadable] [unreadable] The completion of the sequences of whole genomes provides the foundation for determining the structure and function of entire sets of gene products. The structural genomics effort will depend, like the sequencing effort, heavily on the development of high-throughput technologies in many areas. Gene expression, crystal growth, data collection, structure solution, structure refinement, and deposition into structural databases are the necessary steps when X-ray diffraction is used to determine the macromolecular structures. This proposal concentrates on development of technology for the structure refinement step. Because of this work, the data deposited in the structural databases will be improved both qualitatively and quantitatively. [unreadable] [unreadable] High throughput implies that (1) the process is highly automated; (2) the process scales linearly (or better) with investment in hardware and software, and (3) the process scales much better than linearly with investment in personnel. Procedures that require direct, rate-limiting attention of highly trained personnel cannot be considered high throughput. Refinement of macromolecular crystal structures presently requires the intervention of trained crystallographers at several steps, but most especially at those steps which require revision of the model. [unreadable] [unreadable]